Blind Image Quality Assessment of Screen Content Images via Fisher Vector Coding

Abstract

This paper presents an effective blind image quality assessment (BIQA) method for screen content images (SCIs) based on Fisher vector encoding, with the hypothesis that local statistics will be altered with the change of distortions, and can be characterized by the fusion of statistical models and direction vectors. Firstly, a specific Gaussian mixture model (GMM) is generated from a corpus of pristine SCIs to simulate the local distribution of SCIs in spatial domain. Then, discriminative features are generated to characterize the quality of test image with Fisher vector coding and generated GMM. Finally, support vector regression is adopted to learn the mapping between discriminative features and subjective opinion scores. To validate the performance of our method, extensive experiments are conducted on three public SCI databases and the results well confirm its superiority over the existing relevant BIQA method of SCIs.

Keywords

• Blind image quality assessment
• screen content image
• Gaussian mixture model
• fisher vector coding